Two taps before the data hits the ERP

The problem

The fear is that AI writes into the ERP on its own; the reality is that nobody wants to type.

In a plant under this much quality pressure, the strongest cultural pushback to digitalization is the fear of AI "entering data on its own" and contaminating the master ERP — but the supervisor doesn't want to type either.

• In a battery plant under OEM and IATF pressure, the strongest pushback to digitalization is fear of an AI entering data on its own and contaminating the master ERP. One wrong value in the master data can propagate into every report an OEM later receives.
• At the same time the formation supervisor does not want to type: he is managing acid gravity, charge circuits and lot changes, not keyboards. Every keystroke he spends is attention taken away from a process that does not forgive distraction.
• The result is one of two bad outcomes: data entered without anyone checking it, or data never captured at all out of distrust. Either way, the ERP stops being a source anyone trusts for formation data.
• When an OEM asks who validated a formation lot and on what evidence, neither outcome produces a name. And without that name, the answer to the customer becomes a promise instead of evidence.

How it fits the IRIS system

Connect with early human verification — two taps at the formation line.

Connect early human verification mode. Every capture (photo, voice, parsed email) first reaches a tablet with a 2-second visual summary. The supervisor confirms (✓) or corrects (✏️); only then does the data cross into the ERP.

This is the piece that separates iLEAN from an autonomous agent: the AI assembles, a person confirms. Without it, none of the other eleven cases gets approved by quality, because nobody signs off on AI writing into the system a warranty claim will be checked against. Two taps is the price of that trust, and supervisors accept it because it replaces typing rather than adding to it.

See the full IRIS architecture →

Before and after

Formation data entry today versus the two-tap gate

Aspect	Today	With iLEAN Connect
How formation data reaches the ERP	Typed later, or never	Confirmed in seconds at the line
Who is recorded as validating it	Whoever typed it	The supervisor who saw the summary
Supervisor's effort per capture	Walk to a terminal and type	A 2-second summary and two taps
AI writing into the master ERP	The fear that blocks the project	Impossible without a human tap
Corrections to a captured value	Lost in an email thread	Stored with author and reason
Answer to an OEM audit question	A checkbox	A name, a time and the evidence seen

Unvalidated data contaminating the ERP, or data never captured out of distrust → data captured and validated within seconds, clean master data.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• No standalone payback: it is the enabling piece for the whole matrix, with strategic value that is not directly monetizable.
• It is what lets quality approve every other case, because it guarantees clean master data in the ERP.
• Data captured and validated within seconds instead of typed days later or never captured.
• And every value in the ERP carries the name of the person who confirmed it, which is what an OEM asks for after a claim.

The enabling piece for the whole iLEAN matrix — strategic value, not directly monetizable.

And the fair question from the production manager

“If a person confirms everything, what is the AI actually saving?” — the work is not in deciding, it is in assembling. Reading a photo, a dictation or an email into fields is an anchored task where the best models drop below 1.5% error [1], so what reaches the tablet is already right almost every time. The supervisor only confirms or corrects, and every correction is stored so extraction improves at that exact point.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.